Latest / Connor with Honor: AI, Real Estate and Real Life / AI vs Machine Learning vs Deep Learning, Explained Simply

Transcript

AI vs Machine Learning vs Deep Learning, Explained Simply · Connor with Honor: AI, Real Estate and Real Life

• 0:00
  SPEAKER_00: Well, there's a lot to learn.
• 0:01
  SPEAKER_00: This is going to be number two.
• 0:03
  SPEAKER_00: What's the second thing?
• 0:04
  SPEAKER_00: So here we're at Santa Clarita Artificial Intelligence forward slash blog.
• 0:08
  SPEAKER_00: I did a class yesterday talking about question number one, the foundations of AI, what is it, how it works.
• 0:14
  SPEAKER_00: And again, just something that I believe that I'm able to wrap my mind around.
• 0:19
  SPEAKER_00: Hopefully, if you're into this world, or you should be, you should at least be figuring it out.
• 0:25
  SPEAKER_00: We're going to cover section two today.
• 0:26
  SPEAKER_00: So you can see that you can always, I'm going to go through 30 of these things by the time we get done.
• 0:32
  SPEAKER_00: So please feel free to check them out so you can stay abreast of what's happening.
• 0:37
  SPEAKER_00: Question number two: what is the difference between artificial intelligence, machine learning, and deep learning?
• 0:41
  SPEAKER_00: And now there's also agentic learning or systems that are basically like Alpha Evolve.
• 0:46
  SPEAKER_00: It's teaching itself by having particular agents that have been designed, synthetic, artificial intelligence type that work together and actually make their code better, make the machine better by just asking questions.
• 1:00
  SPEAKER_00: How can we get improved performance?
• 1:02
  SPEAKER_00: How can we make our responses more accurate?
• 1:05
  SPEAKER_00: Those types of things.
• 1:06
  SPEAKER_00: And it's a whole big deal.
• 1:08
  SPEAKER_00: We are no longer on the, you know, setting it and having human beings work on it.
• 1:14
  SPEAKER_00: Now it's almost like a rotisserie oven.
• 1:17
  SPEAKER_00: And you kind of set it and you forget it.
• 1:19
  SPEAKER_00: Hopefully, we do have people watching.
• 1:20
  SPEAKER_00: So, question number two: what's the difference between artificial intelligence, machine learning, and deep learning?
• 1:25
  SPEAKER_00: There's three things.
• 1:26
  SPEAKER_00: So, one of the references that people like to talk about, if you've looked at Russian nesting dolls, remember the dolls, and you you take off one, it's got another one, they're egg-shaped, right?
• 1:36
  SPEAKER_00: And then there's like 50 in there.
• 1:38
  SPEAKER_00: Basically, it's kind of a trichotomous thing where you have the overall outside, which is going to be AI.
• 1:44
  SPEAKER_00: That's the overall system.
• 1:46
  SPEAKER_00: And then from there, me, machine learning is a subset of artificial intelligence, and then deep learning is even deeper within that nesting doll platform or that nesting doll theory.
• 1:57
  SPEAKER_00: But what ends up happening is artificial intelligence, that's the broadest concept.
• 2:02
  SPEAKER_00: It's the overarching goal of creating machines that can simulate human intelligence.
• 2:06
  SPEAKER_00: Now, this includes any technique that enables computers to mimic human cognitive functions like problem solving, learning, planning, reasoning, perception.
• 2:13
  SPEAKER_00: Early AI relied heavily on rule-based systems and logical deduction.
• 2:17
  SPEAKER_00: Modern AI heavily leverages machine learning.
• 2:20
  SPEAKER_00: You've probably seen this.
• 2:21
  SPEAKER_00: You go into a Chat GPT or a GROK or one of the different systems and you ask it a question.
• 2:26
  SPEAKER_00: A lot of the times you understand that it has all the data, and sometimes it's biased.
• 2:30
  SPEAKER_00: Sometimes it does give information that's a little askew from our general understanding.
• 2:35
  SPEAKER_00: If you look at large language models that are built elsewhere, like Deep Sea through China, you're also going to get some restrictive responses, which might be a little bit different than the American models.
• 2:46
  SPEAKER_00: Again, it's the data set that these systems are educated on.
• 2:50
  SPEAKER_00: And that's where you're going to get these biases and in some cases hallucinations, which would be if you're asking for something, it's giving you something else or something wrong that could be deemed a hallucination.
• 3:01
  SPEAKER_00: Maybe it's the bad data it was trained on, or maybe it's the output.
• 3:04
  SPEAKER_00: But again, you they're getting much, much better.
• 3:07
  SPEAKER_00: But you still see that.
• 3:08
  SPEAKER_00: Machine learning, it's the subset of AI, so it's the inner, the innermost, not almost the innermost all.
• 3:14
  SPEAKER_00: So instead of explicitly programming a computer for every possible scenario, ML focuses on developing algorithms that allow computers to learn from data without being explicitly programmed.
• 3:24
  SPEAKER_00: So instead of telling it, you know, here's a here's a block, and I need you to put the round peg and the round hole and the square peg and the square hole and the triangular pig into the triangular hole and so on, you basically give it this, like we've done with chess, right?
• 3:42
  SPEAKER_00: You teach it the game of chess, you give it all the rules to jest and then chess, and then you let it figure itself out.
• 3:48
  SPEAKER_00: That type of machine learning that does much better than one that we've actually showed it how to play the game.
• 3:54
  SPEAKER_00: That's on all different platforms.
• 3:56
  SPEAKER_00: Same thing.
• 3:57
  SPEAKER_00: If you have the medical field, you have a doctor that teaches AI about medicine, it doesn't do as well as when you give AI the rule book or all of the particulars with regard to the entire medicine world, and then it comes up with some very nouveau and great ideas regarding how to best approach medicine.
• 4:16
  SPEAKER_00: Same thing with mathematics and physics, it's creating new stuff, it's solving these things, which has gone beyond the human capability, at least at this point.
• 4:26
  SPEAKER_00: You look at people from the past like Einstein, just incredibly sharp, right?
• 4:31
  SPEAKER_00: He came up with some phenomenal things that even kind of brought into question some of the Newtonian physics that people put together.
• 4:37
  SPEAKER_00: That being said, we now have something that has been created that's even smarter than us, that's going to be blowing through those particular statistics and those goals that we've had in the past.
• 4:48
  SPEAKER_00: We're gonna just blow past them.
• 4:49
  SPEAKER_00: So that's where machine learning comes in.
• 4:51
  SPEAKER_00: So common machine learning techniques include supervised learning where models learn from labeled data, labeled data, labeled, and unsupervised learning where it finds patterns in the unlabeled data.
• 5:01
  SPEAKER_00: Then reinforcement learning where an agent learns by interacting with an envoy environment and receives rewards or penalties at the very beginning of artificial intelligence, and probably in some of the training ones now, you have multiple agents that are given a particular task to, let's say, identify a number, a particular number, let's say a six.
• 5:18
  SPEAKER_00: The ones that don't get it, they get destroyed, they get deleted.
• 5:21
  SPEAKER_00: It's gonna be interesting to see if there's any carryover from that when the systems become even smarter, more intelligent, and well, are given more power by the human beings that at least at some point had control.
• 5:33
  SPEAKER_00: Hopefully they still do.
• 5:34
  SPEAKER_00: I think with all the disruption that's gonna be caused by AI, it's not gonna be AI itself.
• 5:39
  SPEAKER_00: It's the human beings that are controlling the AI.
• 5:41
  SPEAKER_00: So that's something else to consider.
• 5:43
  SPEAKER_00: If it's something you think needs to be stopped or controlled, or some kind of uh barrier put in, or some kind of controls or safety mechanism, you might want to talk to your politicians if that happens to be the thing.
• 5:55
  SPEAKER_00: I would definitely learn a little bit more about that if I were you, if you're not.
• 5:59
  SPEAKER_00: Finally, the innermost doll is the deep learning.
• 6:02
  SPEAKER_00: It's a subset of machine learning.
• 6:04
  SPEAKER_00: Along with that is going to be the autonomic type learning as well.
• 6:07
  SPEAKER_00: And these are systems that are basically improving on their own.
• 6:12
  SPEAKER_00: They're giving rules, they're giving tasks, and they're starting to develop their own rules, their own tasks, their own goals, and they're moving forward.
• 6:18
  SPEAKER_00: They're making themselves better in every way possible, going way beyond what us as human beings are capable of.
• 6:24
  SPEAKER_00: And in fact, the way that they communicate between agents no longer under is understood by a lot of people that are watching this because it's coming up with its own method of communication, where maybe as I'm speaking in this video, I might say, I don't know, 1200, 1300 words.
• 6:38
  SPEAKER_00: Uh AI can do that in a much more conservative way because uh the computer is actually much more conservative than us.
• 6:46
  SPEAKER_00: So it's gonna be able to convey a much deeper inference of thoughts than we ever would be able to in the way that we just speak these particular words.
• 6:57
  SPEAKER_00: So it's gonna be much deeper.
• 6:58
  SPEAKER_00: If you look at the Greek language, in some cases, some of the words in Greek takes they take paragraphs of English to really understand and and and grasp, that's the Koine Greek language, so you have a superior in in somewhat ways that's gonna be very similar to the the different ways that AI is gonna be communicating with itself.
• 7:16
  SPEAKER_00: We'll have to keep an eye.
• 7:18
  SPEAKER_00: Um we also have output layers in the neural network.
• 7:21
  SPEAKER_00: More layers allow the network to learn more abstract and hierarchical representations of data.
• 7:27
  SPEAKER_00: For instance, an image recognition, an early layer might detect edges, middle layer might combine edges to form shapes, a deeper layer might combine shapes to identify objects like faces or cars.
• 7:38
  SPEAKER_00: Deep learning has been responsible for many of the most impressive AI breakthroughs in recent years, including highly accurate image recognition, natural language translation, and autonomous driving systems.
• 7:52
  SPEAKER_00: The self-driving cars hasn't been new.
• 7:55
  SPEAKER_00: This has been around since 2010-2012, and you can thank Google for building a lot of that stuff into the world, also coming up with the Transformer architecture that these particular systems use.
• 8:06
  SPEAKER_00: Again, Google, and Google doesn't seem to be the system out there that is creating all the buzz and all the drama.
• 8:15
  SPEAKER_00: They seem to be just they got their nose to the grind the grinder and they're moving forward.
• 8:20
  SPEAKER_00: They're keeping their mouth shut and they're developing these sending out these brilliant research papers.
• 8:25
  SPEAKER_00: I think Google's the sleeper.
• 8:26
  SPEAKER_00: I think we need to watch and see what happens.
• 8:29
  SPEAKER_00: So, in summary, you got AI, that's the big goal, making machines intelligent.
• 8:33
  SPEAKER_00: You got ML machine learning, a way to achieve AI and machines learn from data, and then of course DL, the deep learning, a specific, powerful method within ML using deep neural networks to learn complex patterns.
• 8:46
  SPEAKER_00: Again, simply just kind of a breakdown of how the systems work.
• 8:50
  SPEAKER_00: Next video, we're going to talk about how AI learn exactly and give you an idea and a glimpse into that particular world.
• 8:57
  SPEAKER_00: I'm Connor with Honor, Santa Clarita Artificial Intelligence.
• 8:59
  SPEAKER_00: We'll see you in the next one.
• 9:00
  SPEAKER_00: Thanks for watching.